Download Marrying Customer Intelligence with Customer Segmentations to Drive Improvements to Your CRM Strategy

Paper CI-03-2013
Marrying Customer Intelligence with Customer Segmentations to Drive Improvements to Your CRM Strategy
Phil Krauskopf
Elite Technology Solutions
Willoughby, OH
Abstract
There are a plethora of segmentation schema available that can be used to group your customer base. However,
choosing which one is best for your customer base is often a point of differing opinion within a marketing department. And
once a particular schema is chosen, selecting which intelligence-gathering activities should be applied in order to learn about
the customers within each segment presents its own set of issues. Finally, integrating that intelligence into your current
marketing strategy to create a structured and comprehensive, even stronger strategy muddies things even further.
This paper reviews some of the most common segmentation schema which the author has encountered in Industry,
and some ideas on how to select among various options. He will also discuss some of the intelligence-gathering tools
available to the data scientist, and how this intelligence might be integrated into a marketing strategy.
Introduction
Over the years countless analytical resources have been expended trying to understand customer psyche and
behavior, in pursuit of that magical one-to-one marketing tactic which would allow marketers to communicate with each
individual customer uniquely, based on that customer’s individual needs, wants and motivations. And while it is unclear
whether this goal can ever be achieved, it is clear that customers come in distinct “flavors” that can help marketers refine the
marketing tactics they employ.
There are a plethora of segmentation schema available to group a business’ customer base. However, choosing which one is
best for a particular customer base is often a point of contention within a marketing department. And once a particular schema
is chosen, selecting which intelligence-gathering activities should be applied in order to learn about the customers within each
segment presents its own set of issues. Finally, integrating that intelligence into your current marketing strategy to create a
structured and comprehensive, even stronger strategy muddies things even further.
This paper will review some of the most common segmentation schema which the author has encountered in Industry, and
some ideas on how to select among various options. It will also discuss some of the intelligence-gathering tools available to
the data scientist, and how this intelligence might be integrated into a single, structured marketing strategy.
Why Segment Customers?
Customers in general are extremely complex. Each is driven to buy according to his or her own unique wants and
needs, and will respond to specific communications. Indeed, this is the whole concept behind the idea of so-called “one-toOne” marketing. And while tailoring marketing tactics at an individual customer level may be impractical , we can often find a
relatively small set of tactics that can be profitably applied to a large majority of a customer base.
The human mind wants to categorize everything it “sees.” Assignment into a particular category allows a mind to infer certain
characteristics to members of that category, and to conceptualize and plan interactions with that group using these inferred
characteristics. Indeed, even the name assigned to a particular category can subsume characteristics to that group- whether
or not those characteristics are intended or even true. By grouping customers into segments marketers can begin to
understand these segments, and develop differentiated marketing tactics intended to appeal more strongly to many folks in a
particular segment than would a less targeted tactic (e.g., Marketing shampoo to someone with hair is much more likely to be
successful when marketing to someone with hair, than to someone without). The results of employing these differentiated
marketing tactics can be astounding, improving the return of marketing efforts many-fold.
What is a Customer?
In theory, understanding what constitutes a customer is fairly straight-forward: each customer is a unique individual,
differentiable from other humans via DNA comparison, fingerprint comparison, voiceprint comparison, etc. Unfortunately when
a customer comes into shop, or clicks the Checkout button on-screen, these identification techniques are unavailable. In fact,
name and address information is often unavailable. Instead, customer identification is often inferred using less accurate
means: tender information, phone number, loyalty number, email address, etc. Businesses try to piece together these
disparate pieces of information to obtain a quasi-unified view of purchasing patterns for each customer. And while many
algorithms perform relatively well at piecing together these pieces of information, a marketing strategist should keep in mind
1
the imperfections in such an approach (products intended for a particular demographic, say young adults, may appear to be
purchased by older ones, simply because parents are providing the tender for transactions).
Characteristics of a Successful Schema
It’s critical to understand that there is no single, “magical” segmentation schema for any business. Indeed there are a
plethora of ways to categorize customers. And while there are many analytical tools that may help understand customers, the
“best” segmentation is the one that dovetails with, and helps drive, a particular marketing strategy accepted by the P&L
manager.
In fact, segmentation schema are usually built in conjunction with the strategic planning process, and are often informed by the
financial objectives given the P&L manager. P&L managers are given objectives like “increase sales by x%,” or “increase
profits by y%.” In turn, the P&L manager develops an initial strategy based on a meta-segmentation directly connected to the
P&L model (“Lessee, if we increase customer retention by this much, increase new customers by that much, and increase
spend of existing active customers by this other amount, then the P&L model indicates that we’ll meet our financial
objectives”).
Aside from this need to conform to the P&L Manager’s conceptualization of the customer portfolio, based on my experience
there are three additional characteristics necessary for a schema to be successful. A successful schema must be
Identifiable
Measurable
Actionable
In order for a schema to be Identifiable there must be a clear set of customer characteristics which objectively classify a
customer into a particular segment. Moreover, the information needed to evaluate these criteria must be readily available at a
customer level, so that the segmentation can be projected onto the database. “If you can’t find them you can’t market to
them.”
A schema is Measurable just when its performance can be measured, not just at an individual response level, but how that
performance impacts the profitability of customers in its segment, and how each segment contributes to overall profitability of
the business. Thus a measurable schema provides a complete integration of its segments with the associated P&L model,
and can be used to identify where opportunities lie.
Finally. A schema is Actionable just when differentiated marketing tactics can be developed for each segment within a schema,
as well as executed against each segment. The only reason for separating out a particular group of customers is if there is an
intention, or potential intention, to employ differentiated marketing tactics against that segment. To do this requires an ability to
communicate directly to members of a segment, and to provide them an offer designed specifically for members of that segment.
Customer Portfolio
Segment Identification
Marketing
Tactics
Customer Segment
Figure 1: Using Customer Segmentation to Drive
Profitability
Common Customer Segmentation Schema
Below are several of the most common customer segmentation schema I have run across in Industry. Most
segmentations begin as some form of a Customer Lifecycle segmentation (below), and then overlaid with additional customer
segments. The schema discussed compose, by no means, an all-inclusive sampling of customer segmentation schema,
which vary as much as the businesses themselves. They do, however, cover the most common types I’ve seen.
2
Customer Lifecycle Segmentation
The most common basic segmentation schema is the Customer Lifecycle Segmentation (CLS), chosen perhaps
because it is easy to conceptualize, and lends itself quite well to the strategic planning process. This schema is based on the
assumption that customers follow characteristic behaviors as they transition through their relationship with a business, from
new customer through attrition. Figure 2 shows a typical CLS.
Active
New
Customers
Lapsed
Deeply
Lapsed
One-Timers
Figure 2: Typical Life-Cycle Segmentation Schema
The Prospects segment comprises those folks who are not currently considered customers. This will include folks who have
never made a purchase, and may also include customers who have lapsed so deeply that they effectively “act” like noncustomers. Offering tactics to this segment are designed to bring the prospect in the door, and may sacrifice short-term return
for a longer-term relationship. Audience Selection models, or additional sub-segmentations, of this group may be employed, in
order to boost response rates to an acceptable level.
The New Customer segment constitutes those Prospects who have recently made their first purchase. Businesses often
employ a nurturing, or on-boarding strategy for new customers, offering them additional information about the business’
product lines, or communications about product pricing, in an effort to further engage the customer. Offers may be a bit more
lucrative than those received by tenured customers, but not nearly as rich as those sent to Prospects. Because little
information is available about New Customers, additional segmentations are not usually deployed against them, although
audience selection modeling may be deployed in order to focus on more promising customers.
A customer remains in the New Customer segment for a particular period of time, after which a new customer becomes
unacceptably unlikely to make a second purchase before lapsing. The exact period of time is usually based on a survival
analysis of new, non-repeat customers, or an analysis determining when the behavior of repeat new customers begins to
coalesce with the behavior of repeat tenured customers.
Those who fail to make that second purchase prior to reaching that selected level of tenure are placed into their own segment
(One Timers) and treated separately. Well over 50% of New Customers often migrate into this segment, so this group can
represent a significant opportunity cost to a business. Understanding the motivations of this segment is critical to the
development of successful marketing tactics for it. Formal surveys combined with observational/anecdotal information, can
provide marketing strategists with a rich source of ideas for successful tactics.
Those New Customers who are successfully lured into making multiple transactions are placed into a category called Active
Customers. This segment includes all tenured customers who have made a “recent” purchase (the level of recency varies by
industry, but is often twelve months). Because of the quantity and accuracy of information available for these customers, the
motivational variations within this segment, and the opportunity costs associated with this segment, businesses often expend
considerable analytical resources studying and further sub-dividing this segment, in order to refine marketing tactics.
A customer who fails to make a purchase for a specified period of time (often one year) is placed into the Lapsed Customer
segment. This is a mixed bag of customers— some have only ever made a single purchase, while others were once prime
contributors to a Business’ bottom-line— with just as many reasons for halting purchasing. While this group tends to be more
3
responsive to marketing tactics than other segments like Prospects, New, and One-Timer segments, careful surveying into this
segment can identify motivational sub-segments for which better-targeted tactics can be developed. And audience selection
modeling can improve results even further.
Customers who lapse even further (often 2+ years) have confirmed their intentions to no longer transact with a business.
While this segment still tends to be more responsive to marketing tactics than do Prospects, many businesses treat this
segment as just another source of Prospects.
Segmentations based on KPI Drivers
As I mentioned earlier the Active Customer segment is often subdivided even further for marketing purposes. Some
of the most effective sub-segmentations I’ve run across are those built using a business’ Key Performance Indicators (KPIs).
Clustering and CART-CHAID techniques can be used to group customers according to similar KPI values, which can often be
used to inform the design of marketing tactics (increase the number of trips for this segment; increase the margin rate for that
segment, etc). And because the number of segments identified by clustering algorithms can be controlled, the cost of
executing these multiple tactics can also be controlled. Moreover, because this segmentation schema ties directly to the
corporate P&L model, forecasting and monitoring are straight-forward.
Segmentations based on Product
Certainly a primary motivator of customer purchasing is product. So understandably the richest source of purchasers
of a particular product are those customers who have purchased the product before; in fact prior product purchasing behavior
can also predict which additional or new products a customer is likely to purchase. And because margin rates can vary widely
across product lines, businesses are strongly incented to encourage purchases of some products more strongly than others.
Consequently product-based segmentations are extensively used by businesses. The structure f this type of
segmentation schema can vary greatly. There are simple product-based segmentations, wherein inclusion/exclusion in a
product segment is based on historical purchasing. Other segmentations are more sophisticated, segmenting customers into
“super groups,” based on associative or directed clustering techniques. Still others may group customers according to
propensity to expand their purchasing into particular product group.
Segmentations Based on Customer Value
A natural way for a business to think in terms of customer relationship is to base the system of offers/rewards on the
value a customer may return to the business. In other words, the return on a customer has to justify the investment. In this
type of segmentation customer value is estimated, either from an historical perspective or in a predictive manner (potential
profitability), and then these values are quantiled. Those in the highest value quantiles receive more communication and/or
richer offers, while those in the lower value quantiles may be excluded from most or all communication; and the offers
communicated to this segment may be much less attractive—at least from a real-value perspective. While this type of
segmentation classifies customers by (potential) return, it doesn’t speak to customer motivation for purchasing. Motivations
may be the subject of additional segmentation overlays.
Psychographic Segmentations
Prior purchasing history almost always overwhelms all other candidate predictors when trying to predict customer
behavior. However, there are some segments where we have little, if any, purchasing behavior with which to predict behavior.
Indeed, there is a dearth of prior purchasing history available for Prospects, New, and Deeply-Lapsed customers. And recent
purchasing history for even those customers who are simply Lapsed may be inadequate to predict customer behavior. In
these cases psychographic segmentations may be employed to refine marketing tactics.
A psychographic segmentation attempts to group customers according to similarity in personality, values, attitudes, interests,
and lifestyles, in the belief that those folks who share these characteristics are likely to react similarly to a particular marketing
tactic. These segmentations are usually based on extensive market research and/or demographic data. Segments may be
defined using anecdotal observations of the data, or they may be defined using more sophisticated clustering techniques. I’ve
seen both techniques work for the segments mentioned above, although over-sophistication can often be a problem. Note too
that there are several sources of commercially-available psychographic segmentation schema.
Segmentations Based on Customer Offer Preferences
Some customers live for the “good deal,” while others respond best to “contest” types of offers. Still others want
access to products and information not available to the “normal” customers. And most businesses have a core of customers
for whom it would take a stick to keep them away. Each offer preference type needs its own method of communication, and its
own offering strategy, in order to maximize the success of marketing tactics.
4
A business’ active customers will often have enough purchasing history to allow development of multinomial models to
categorize customers according to offering tactic preference. Model design is critical here. One must ensure that the
preferences being modeled are different enough to provide good differentiation, and be actionable at the same time. Utility
Maximization and Choice Modeling are two modeling techniques that may be successful in modeling motivations. Testing can
determine model cutoffs for inclusion into each preference category.
Customer Intelligence Gathering
Categorizing customers into groups is only the start of a good marketing strategy. Understanding the behavioral and
motivational commonalities of customers within each segment is just as important. The toolbox of a marketing analyst must
indeed be large in order to hold all the analytical tools available. Otherwise an analyst could easily fall prey to the Law of the
Hammer (“If all you have is a hammer then everything looks like a Nail”). Following are brief descriptions of the most common
intelligence-gathering tools I’ve run across. As with the segmentation schema discussed above, this is by no means a
complete list, although it does encompass those I’ve seen most successfully employed.
Migration Analyses (Customer Churn)
Segmentation schema are seldom static. Customers flow between segments, both organically, as well as a
consequence of marketing strategy. A migration analysis is similar to an attrition analysis, but on a grander scale. While an
attrition analysis concerns itself with threat of attrition, a migration analysis tries to capture flows between all segments, not just
purchasing cessation. Thus customers flowing from more to less profitable segments (or vice-versa) are also investigated.
Migration seasonality is often considered, to identify seasons when a customer or segment is most susceptible to migrationaltering tactics. Historical trends (vintage analyses) are also important, in order to identify trends early enough to react
successfully to those trends. While migration analyses are often delivered as standard reports (Vintage report, Cohort Report,
Migration Report, etc.), the analytic team should be responsible for interpreting these reports and, more importantly,
investigating further any trends which arise.
Customer Basket Analyses
Similar to a Market Basket Analysis, a Customer Basket Analysis is more longitudinal in nature, concerning itself with
more than just a single customer transaction. Like a market basket analysis, this type of analysis looks at the selection of
products a customer purchases, but folds in a temporal component as well. Thus it can better identify true Riders and Drivers
across categories, as well as product migration patterns. New metrics are also available. The concept of Product
Equivalence, where a customer is relatively agnostic when choosing between a set of products, can help marketers better
understand the likely impact of a product-dependent tactic. Similarly this temporal aspect allows a deeper analysis of product
migration (Next-Logical-Product types of analyses).
Seasonal/Annual Holdouts
Marketing can be viewed as an investment intended to increase profitability. While campaign-level holdouts are
somewhat common, the total return on marketing dollars can’t be calculated simply by summing up all the campaign
incrementals (interactions between various marketing activities belie this approach). To better measure return on marketing
dollars some businesses employ the use of seasonal and/or annual holdouts. This method entails the imposition of a formal
experimental structure onto the customer data mart in such a way as to allow incremental measurements across a season or a
year. Additional structure can be employed to allow other measurements, such as incremental performance by purchase
channel, or by marketing channel. If the analysis investigates incrementals at the level of KPI-driver, one can determine which
KPIs are driving results, and thereby provide insights as to how to improve the results. Most segments can be defined in a
relatively straight-forward manner, but any segments containing customers new to the business over the course of the
season/year require special consideration.
Demographic Profiling
Marketing to prospects is notoriously problematic, both because of the low conversion rates, as well as the difficulty
of developing these new customers into lucrative ones. One approach to finding “good” new customers is to look for ones who
are similar to existing “good” customers. This is often done by choosing prospects whose population statistics are similar to
those of a business’ “best” customers. While most demographic profiling is performed on a limited set of population statistics
(because of the extensive confounding between these statistics), it is possible to use more advanced methods (factor analysis
and Principal Component Analysis) to reduce the confounding effects and thereby obtain a more refined view of these profiles.
Customer Surveys
While observationally-based analytics can reveal much about customers, one thing they can never do is reveal why
customers behave the way they do. Understanding these whys, though, is the key to developing the most successful
marketing tactics. And there’s seldom a better way to gain insight into these reasons than to simply ask the customer directly.
Even though customer surveys come with their own set of caveats, they can also be a remarkable source of candidate
5
marketing tactics, even when only treating these responses anecdotally. Advanced quantitative methods may refine the list of
customer motivators here, but it’s important to remember that any tactics developed from surveys need to be tested prior to
productionization.
Applying Structured Marketing Tactics to Your Customer Segments
As mentioned earlier, the only reason for a segment to exist is to develop differentiated marketing tactics for that
segment. So when developing marketing tactics these tactics should align with the segmentation schema deployed against
the customer base. Generally a boot-strapping approach, or Champion-Challenger strategy is used, incrementally improving
performance on each segment.
The first step in developing a successful marketing strategy for a segment is to find an initial tactic that can serve as a
basis of comparison for future improvements. One way of doing this is to collect and review all the intelligence gathered on
the segment, and use a brain-storming session to develop several candidate tactics. Then design a comparative test using a
relatively straight-forward A-B Splits design within the segment.
The number of tactics that can be tested at once will be limited by the amount of circulation available for testing, and the
desired resolution of the test. Using adequate sample sizes is a fundamental requirement of successful testing, and is
probably the most common reason I’ve seen for unsuccessful test execution. Adequate sample sizes can only be determined
by first, understanding the critical success measures for a particular test, as well as determining some kind of estimation of the
test margin of error, given the selected sample size. One also needs to understand how small the margin of error must be in
order to differentiate between two tactics. Finally, the margins of error of the results must be incorporated into the analysis.
Once an initial “best performing” tactic is found, a Champion-Challenger strategy can be deployed against a segment to
provide a structured approach to incremental improvements of the champion strategy. In a Champion-Challenger Strategy
candidate alternative tactics are developed for marketing to a particular segment, and then applied to a simple random sample
of the segment, while the remainder of the segment receives the current “champion” tactic. Assuming the customers receiving
the challenger tactic are statistically identical to those receiving the champion tactic, the results can be analyzed to determine
a “winner.” If the challenger tactic out-performs the champion then (usually after some confirmatory testing) the challenger
becomes the new champion. In any event new challengers are developed, and then evaluated against the current champion
tactic.
Develop new
challenger
tactics
Detailed KPI
analysis
Test
challengers
against current
champion
Assign new
champion (if
appropriate)
Evaluate
performance
Confirmational
testing (if
appropriate)
Figure 3: Champion-Challenger Strategy
6
One of the best sources of new champion tactics can be found by analyzing prior Champion-Challenger tests at a P&L KPI
Driver level. Understanding how each tactic affects these drivers, and how these effects compare to those of other tactics,
one can begin to understand how a particular tactic drives performance, and where it falls short. Other sources of tactical
improvements include market research, economic trends, and other new intelligence gathering activities.
Sometimes even the best tactics don’t provide sufficient ROI to warrant the investment in differentiated marketing necessary to
deploy this tactic. In these cases a common “next step” is to develop an Audience Selection Model to improve returns. An
Audience Selection Model is a predictive model which ranks customers by their expected susceptibility to a particular
marketing tactic. These rankings are then used to cull the segment in a structured fashion, until the ROI is high enough. It is
important to note that these models do not improve performance of individual customers; rather, they increase the overall ROI
of a campaign, at the sacrifice of less profitable customers.
Customers in the lowest performing model ranks are not necessarily less valuable. It may be that other tactics will work for
these customers, or that maybe relationships with these customers need to be more fully developed. In either event, if there
are enough if these lower-ranking customers, it may be worthwhile to segment these folks out separately, and employ this
same Champion-Challenger strategy against them as a distinct new segment.
Conclusion
As evidenced by the contents of this paper the marketing analyst needs a large, diverse selection of analytical tools
available in his or her toolbox for developing a successful marketing strategy for a company’s customer base. And this large
toolbox should be constantly expanding as new marketing tactics and technology becomes available. While many of the “tools”
discussed herein were initially developed for traditional direct mail, most of them are applicable to all the new marketing
technology (SMS, Online, E-mail) available today.
SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in
the USA and other countries. ® indicates USA registration. Other brand and product names are registered trademarks or
trademarks of their respective companies.
Contact Information:
Name: Phil Krauskopf
Enterprise: Elite Technology Solutions
Address: 5900 SOM Center Rd, Ste 12-276
City, State ZIP: Willoughby, Ohio 44094
Work Phone: 440-943-4575 x112
Fax: (440) 943-4822
E-mail: Phil.Krauskopf@Elite-TS.com
Web: http://www.elite-ts.com/
7
8